Kelp forests are vital marine habitats with significant ecological, cultural, and economic importance. These ecosystems, found along coastlines, are susceptible to regional and global stressors including coastal development and climate change. This paper presents Satellite-based Kelp Mapping (SKeMa), a novel framework for automatically mapping canopy-forming kelp forests along the coast of British Columbia using Sentinel-2 satellite scenes. SKeMa employs a deep learning semantic segmentation model, offering an efficient alternative to traditional, labor-intensive, and time-consuming kelp-mapping methods. Here, we initially evaluated SKeMa against two conventional, non-deep-learning baselines, NDVI thresholding and Random Forest, and several alternative deep learning architectures. For the best-performing model, an ablation study on input channels identifies a suitable input configuration, and a cross-validation study with independent test sets yields a mean Intersection over Union (IoU) of 0.6024 and F1-score of 0.7491. Average precision and recall are 0.7261 and 0.7951, respectively. Overall, the highest validation metrics were for images with medium to large kelp beds and the lowest for images with small kelp beds, likely because Sentinel-2’s spatial resolution limits mapping of small kelp beds. These results are achieved under a highly imbalanced class distribution, with labeled kelp pixels constituting, on average, less than 0.1% of all pixels in a Sentinel-2 scene. The results demonstrate the model’s ability to detect kelp canopies across diverse coastal regions, particularly for medium to large kelp beds.